Colgate says LLM role-play can generate ~90% accurate survey data, challenging market research

ZeroStateReflex · x · 2026-11-22

Colgate published a paper showing LLMs can predict real purchase intent by role-playing target demographics, yielding survey data with roughly 90% accuracy — a quiet threat to the market research industry.

The key is Semantic Similarity Rating (SSR): asking an AI to rate a product 1-5 yields safe, middle-of-the-road answers, so researchers instead give the model a demographic profile and a product concept, have it write raw, unfiltered thoughts, then use a semantic model to convert those texts into a rating distribution. Observers note profound short-term implications and disturbing long-term ones.

Original post →

More from AGI Musings

AGI Musings channel →